Design For Early-Warning Model In Chinese Banking Sector

The early-warning system is one of the important issues in bank supervision. It helps bank supervisors to identify the changes in bank condition so that to schedule on-site bank examinations and allocate more efficiently the scarce supervision resources. However, up to now the researches in this field are based on the empirical studies on American commercial banks because there are many failed bank samples and the Fed has a CAMELS rating system that gauges bank conditions. For countries such as China that do not have failed bank samples or a bank rating system we must find a new way to develop early-warning models.This dissertation intends to develop an early-warning econometric model based on the off-site surveillance data. Firstly it applies principle component analyses method to evaluate the condition of banks. It defines a bank as a problem bank if its weighed component score is lower than the average, otherwise as a healthy bank. Thus it builds up a high-risk bank sample and a low-risk bank sample. Then it tests the means of each of the various financial ratios of these two samples to find the financial characteristics of these two groups. It further uses partial correlation analyses to filter out five financial ratios to be the explain variable candidates. By logit regression it finally develops an early-warning model with three explain variables. Both the in-sample test and jackkm'fe test shows that this model is robust with relatively high sensitivity, specificity and overall classification accuracy.